On Mobility Load Balancing for LTE Systems

of 5
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Document Description
On Mobility Load Balancing for LTE Systems
Document Share
Document Tags
Document Transcript
  On Mobility Load Balancing for LTE Systems Raymond Kwan 1 , Rob Arnott 1 , Robert Paterson 1 , Riccardo Trivisonno 1 , Mitsuhiro Kubota 2 1 Telecom Modus Ltd, UK  2  NEC Corporation, Tokyo, Japanraymond.kwan@eu.nec.com  Abstract   — In this paper we present simulation results todemonstrate that a simple distributed intra-frequency loadbalancing algorithm based on automatic adjustment of handoverthresholds can significantly reduce the call blocking rate andincrease cell-edge throughput in an LTE network.  Keywords- LTE, load balancing, handover, self-organising network (SON), radio resource management (RRM) I.I  NTRODUCTION The term ‘load balancing’ can be used to describe anymechanism whereby highly loaded cells distribute some of their traffic to less heavily loaded neighbours in order to makethe use of the radio resource more efficient across the wholenetwork. In this paper, we are concerned with intra-frequencyload balancing mechanisms that have reaction times measuredin several minutes or hours, and which can be implemented in aLong Term Evolution (LTE) network with a minimum of additional signalling.There are many ways to re-distribute load between cells.One approach is to adjust the cell coverage by modifying the pilot power [1]. A larger pilot power effectively allows moredistant mobiles to access the cell, thereby increasing thecoverage area. However, automatic adjustment of cell coveragearea runs the risk of creating coverage holes. Another way tore-distribute the load is to modify the handover regions between neighbouring cells. Such an approach is referred to asmobility load balancing (MLB). The principle behind mobilityload balancing is to adjust the handover regions by biasing thehandover measurements, causing cell-edge users in highlyloaded cells to migrate to less heavily loaded neighbour cells,thereby improving the efficiency of resource utilisation [2].The result is an improvement in call blocking rate and cell-edge throughput. Since the re-distribution of load is doneautomatically between cells, this technique is an example of aSelf-Organizing Network (SON) algorithm.The organisation of this paper is follows. In section II, asimple distributed load balancing algorithm is described. Insection III, a simulation model is described and numericalsimulation results are presented. Finally conclusions are givenin section IV.II.M OBILITY LOAD BALANCING According to [3], a handover can be triggered a number of events. In this paper we are concerned with one particular event, known as event A3, in which it is detected that aneighbouring cell offers a better signal quality for a particular UE than its currently serving cell. This condition can beexpressed as follows. ( ) ( ) η ξ  +++>−++ )()( , csi f iicn ji f  j j OOM OOM  (1)where indices i and  j correspond to the serving and neighbour cells respectively, i M  and  j M  correspond to the UEmeasurement result of cells i and  j , ( )  f i O and ( )  f  j O correspond to the frequency-specific offset for frequency i  f  and  j  f  that are used by cells i and  j respectively, )( csi O isthe cell-specific offset of the serving cell, and )(, cn ji O is the cell-specific offset of the neighbour cell  j with respect to cell i . Thequantities ξ  and η  are a hysteresis term and a fixed offsetrespectively. The measurement i M  may be in the form of theReference Signal Received Power (RSRP) in the unit of dBm,or Reference Signal Received Quality (RSRQ) in the unit of dB[4]. In this paper we consider only intra-frequency handover, inwhich case (1) becomes η ξ  ++−>− )(,)( cn jicsii j OOM M  . (2)From (2), it can be seen that by increasing the offset )(, cn ji O , itis possible to cause a mobile served by cell i to be handed-over to the neighbour cell  j , thereby reducing the load in cell i .In this paper we perform load balancing by automaticallyadjusting the offsets )(, cn ji O based on cell load measurements. Asimple way to do this is as follows ( )( ) °¯°®- <−≥−−Δ− ≥−Δ+ ← th jicn jith jicncn jithi jcncn jicn ji if Oif OOif OOO  ρ  ρ  ρ  ρ  ρ  ρ  ρ  ρ  ρ  ,,,max,,min )(,)(max)(,)(max)(,)(, (3)where Δ is the offset step-size, i  ρ  and  j  ρ  are the load of cell i and cell  j respectively, and the quantity 0 > th  ρ  is a pre-defined threshold for triggering load balancing. All )(, cn ji O areinitialised to zero. The update (3) is applied periodically, inresponse to new load measurements made by each cell. Thusthis method is suitable for a distributed implementation, inwhich each cell exchanges load measurements and offsetvalues with its neighbours.One property of (3) is that the offsets are alwayssymmetrical, i.e. )(,)(, cni jcn ji OO −= . This is important in order to 978-1-4244-3574-6/10/$25.00 ©2010 IEEE  ensure that a UE handed over from one cell to another is nothanded straight back (i.e. to prevent ping-ponging).In order to apply the update in (3) a cell load measurement isrequired. A simple method of measuring load is to calculate themean utilisation of physical resource blocks (PRBs) in the cell,as follows. ¦ Δ= k k i T  K n  ρ  (4)Here k  n is the total number of PRBs allocated to bearer  k  over a measurement period of  T  Δ sub-frames,  K  is the totalnumber of PRBs in the system bandwidth, and the summationis performed over all bearers connected to cell i during themeasurement period. Note that i  ρ  is a value in the range 10 ≤≤ i  ρ  . The problem with this method is that the PRButilisation is not always a good indication of the real load in acell. If there are non-GBR bearers connected to the cell, thePRB utilisation may always look high because the scheduler will assign any ‘spare’ resources in the cell to those users. Evenin the case of GBR traffic, a high PRB utilisation may not be areliable indication of high load because it may be possible toachieve more efficient PRB usage at the expense of slightlyincreased packet delay, whilst still achieving the required QoSfor all connected bearers. We can obtain a more useful loadmeasurement by taking into account the QoS requirements of the connected bearers, as follows. ¦ ¸¸ ¹ ·¨¨© §  Δ= k k reqk k C i T  RT  K n K  )( ,min  ρ  (5)Here k  T  is the achieved throughput of bearer  k  over the period T  Δ and ( ) reqk   R is the required mean data rate for bearer  k  derived from its QoS requirements [5]. (In the case of non-GBR services, ( ) reqk   R represents an ‘equivalent GBR’, i.e. aminimum throughput that would be considered acceptable for non-GBR users). C   K  is a positive real constant (typicallyaround 2.0) used to keep the load under a reasonable value incase k  T  becomes very low due to exceptionally poor radiochannel quality for a particular bearer.III.N UMERICAL R  ESULTS  A.Simulation Model  Simulations are used to evaluate the performance of the proposed load balancing scheme in the downlink of an LTEnetwork. The main simulation parameters are given inTable 1 .We assume video streaming traffic with a constant date rateof 64kbps per user and a mean call duration of 3 minutes.The relatively slow action of load balancing means thatlong simulation durations are needed to study its behaviour.We have simulated a network of 21 cells for a period of 4hours, with the cell offsets )(, cn ji O updated every 60 seconds.For a simulation of this size, fully modeling the MACscheduler, RLC, HARQ and physical layers would be prohibitively complex. Instead we use a simplified (butrepresentative) model as described below. In each sub-frame,an SIR value k  γ   is computed for each user  k  as follows. ¦ ≠ += i jk  jk ik i k   P  N  P  ,0,, α α γ   (6)where 0  N  is the thermal noise spectral density (including UEreceiver noise figure), k  j , α  is the mean path gain (includingshadowing and antenna sectorisation gain but not fast fading) between UE k  and cell  j ,  P  is the transmitted power (assumedto be the same for all cells), i is the index of the serving cell for UE k  , and k i ,  χ  is a unit-mean geometrically distributed randomvariable representing Rayleigh fading between the UE and cell i .The SIR value k  γ   is mapped to a modulation and codingscheme (MCS) based on look-up-tables obtained from link level simulation, to give an instantaneous achievable data rate k  r  . A proportional-fair scheduling metric is then calculated for each user as follows, where  β  =1.5 is a fairness parameter and k  T  is the mean throughput achieved by user  k  .  β   μ  k k k  T r  = (7)In each cell, the connected users are ranked in descendingorder of  k   μ  and each user is assigned as many PRBs asneeded to empty its queue until all users are served or all PRBsare used.An admission control mechanism [6] is included in eachcell which blocks new calls when the measured load reaches0.7, and blocks incoming handover attempts when themeasured load reaches 0.8. A congestion control mechanism[7] is also included which drops users if the load exceeds 0.9.Call arrival is modeled as a Poisson process, and tosimulate the effect of non-uniform geographical trafficdistributions we assume that new calls are generated ingeographically localised ‘clusters’ according to the followingmethod. At any given time there is always the same number of clusters, clust   N  . Each cluster exists for a fixed period, clust  T  ,and when a cluster expires a new cluster replaces it at arandomly chosen location. The call arrival rate of each cluster increases linearly over time from 0 up to a maximum value attime 2 clust  T  , and then decreases linearly to 0 again. The peak call arrival rate is a constant and the same for all clusters. Thelifecycles of the clusters are offset in steps of  clust clust   N T  ,which ensures that at any time the total call arrival rate  summed over all clusters is constant (i.e. the average callarrival rate over the whole network remains constantthroughout the simulation). New UE positions are generated byadding a random displacement to the centre location of thesrcinating cluster. The displacement is generated from a 2Dnormal distribution, and wrap-around is taken into account.This model generates an offered traffic distribution whichcontinually and gradually changes, whilst maintaining aconstant average call arrival rate over the whole network. Therate of change is governed by the parameter  clust  T  , which weset to 2 hours. (After a period of 2 hours, all existing clusterswill have been replaced, so the traffic distribution will havecompletely changed). The degree of traffic localization isdetermined by the cluster ‘radius’, i.e. the r.m.s. distance between new UE locations and the centre of the srcinatingcluster. For illustration, Figure 1 shows the locations of theUEs generated in the first 15 minutes of the simulation for  32 = clust   N  with cluster radii of 25m and 100m.  B.Results In a first set of simulations, the cluster radius is fixed to100m and the simulation is repeated at different mean callarrival rates such that the mean offered load per cell varies between 1 Mbps and 4 Mbps. We compare the cases of noMLB, MLB based on PRB utilisation (equation (4)) and PRB based on load (equation (5)). Figure 2 - Figure 4 show the blocking, dropping, and handover failure probabilities as afunction of the offered load. As expected, as the offered loadincreases, the blocking probability increases. However, load- based MLB provides significant reduction in blocking probability compared to the other cases. The results for PRButilisation-based MLB are poor, confirming the fact that PRButilisation is not a reliable indicator of actual cell load.By biasing the handover decisions users will be handedover when their radio conditions in the target cell are lessfavourable, increasing the risk of dropping. However Figure 3shows that MLB causes little change in dropping probability.(The trade-off between improved blocking and worseneddropping can be adjusted by means of the maximum offset parameter  )(max cn O ) . Figure 4 shows that load-based MLB significantly reduceshandover failure rate at medium load, but slightly increases it athigh load.Figure 5 shows that at medium load, load-based MLByields a significant improvement in 5%-tile user throughput,with little effect on mean user throughput. At both very highand very low load, MLB shows no benefit, because in thesecases the load variation between cells is generally not largeenough to trigger MLB.Figure 6 shows the CDF of cell load aggregated over allcells and over the whole simulation, for the case of meanoffered load 2.5Mbps/cell. As expected, MLB results in anarrower CDF (i.e. more evenly distributed load) but a higher mean load. The mean load is increased because MLB causessome cell-edge users to handover earlier, resulting in worseradio conditions in the target cell, and hence worse averageradio conditions over the whole network. This does not resultin a loss of throughput because it is more than compensated for  by the more even distribution of users between cells.In second experiment, the mean offered load is fixed at 2.5Mbps per cell and the cluster radius is varied from 25m to400m. Figure 7 - Figure 9 show the blocking, dropping, andhandover failure probabilities as a function of the cluster radiusrespectively in the case of no MLB and load-based MLB.These results show that, as expected, the gain of MLBreduces as the cluster size increases, since larger clusters giverise to a more uniform traffic distribution which reduces theload variation between cells. On the other hand, small clustersgive rise to a highly non-uniform traffic distribution, therebycreating larger load differences between cells. In these results,a large gain from MLB is seen even with very small clusters,even though in such cases the gain of MLB is clearly verydependent on the exact locations of the clusters relative to thecell handover boundaries.IV. CONCLUSIONS In this paper, simulations are performed to investigate the performance benefits of MLB in LTE networks. Althoughmore work is needed, the results suggest that significant gainsin call blocking and cell-edge user throughput are possible evenwith a very simple MLB mechanism. However it is also shownthat PRB utilisation is not a sufficient indicator of cell load for MLB, even in the case of constant bit rate traffic.A CKNOWLEDGMENT Telecom Modus Ltd is a wholly owned subsidiary of NECCorporation. The authors would like to thank their colleaguesat Telecom Modus and NEC Japan for their constructivecomments.R  EFERENCES[1]K. A. Ali, H. S. Hassanein, and H. T. Mouftah, Directional CellBreathing Based Reactive Congestion Control in WCDMA Cellular  Networks , in Proc. of   IEEE Symposium on Computers and Communications , July 1-4, 2007.[2]R. Nasri and Z. Altman, “Handover Adaptation for Dynamic LoadBalancing in 3GPP Long Term Evolution Systems”, Proc. of   International Conference on Advancnes in Mobile Computing &Multimedia (MoMM) , Dec. 2007.[3]TS 36.331 3 rd Generation Partnership Project; Technical SpecificationGroup Radio Access Network; Evolved Universal Terrestrial RadioAccess (E-UTRA); Radio Resource Control (RRC); Protocolspecification (Release 8).[4]TS 36.214, 3rd Generation Partnership Project; Technical SpecificationGroup Radio Access Network; Evolved Universal Terrestrial RadioAccess (E-UTRA); Physical layer – Measurements (Release 8).[5]TS 36.413, 3 rd Generation Partnership Project; Technical SpecificationGroup Radio Access Network; Evolved Universal Terrestrial RadioAccess (E-UTRA) S1 Application Protocol (S1AP) (Release 8).[6]R. Kwan, R. Arnott et al., “On Radio Admission Control for LTESystems”,  Proc. of IEEE Vehicular Technology Conference (VTC), Fall  ,2010.[7]R. Kwan, R. Arnott et al., “On Pre-emption and Congestion Control for LTE Systems”,  Proc. of IEEE Vehicular Technology Conference (VTC), Fall  , 2010.  [8]NGMN Alliance,  NGMN Radio Access Performance EvaluationMethodology , January, 2008.Table 1 Simulation ParametersSystem bandwidth 5MHz (25 PRBs)Simulation duration 4 hoursCell Layout Hexagonal grid, 7 cell sites, 3 sectors per site with wrap-aroundInter-Site Distance (ISD) 500mPathloss 128.15 + 37.6log 10 ( max(d km ,0.035) )Shadowing Log-normal with standard deviation 8 dBIntra-site correlation 1.0Inter-site correlation 0.5Antenna pattern (horizontal) ¸¸ ¹ ·¨¨© § ¸¸ ¹ ·¨¨© §  −= mdB  A A ,12min)( 23 θ θ θ  ,where dB20deg,70 3 == mdB  A θ  eNB Tx Antennas 1 per sector UE receive antennas 2Load measurement averaging period T  Δ 400 ms for admission control andcongestion control, further averaged over 60 seconds for MLB )(, cn ji O update rate60 seconds )(max cn O 6 dB Δ 0.5 dBMLB trigger threshold, th  ρ  0.25Scheduler Proportional Fair Traffic model Constant bit rate 64 kbps (64 bytes/8ms)Call duration Geometric with mean 180 seconds ( ) reqi  R 64 kbpsUE speed1 m/s random walk Handover hysteresis 3 dBHandover measurement period100 msAC load threshold 0.7 for new users, 0.8 for handover usersCC trigger load threshold 0.9 clust   N  32 clust  T  2 hours Cluster Radius 25mCluster Radius 100m Figure 1 UE positions generated in first 15 minutes 11.522.533.5402468101214Offered Load Mbps    B   l  o  c   k   i  n  g   P  r  o   b .   (   %   )   No MLBPRB utilisation-based MLBLoad-based MLB Figure 2 Blocking probability as a function of offered load. 11.522.533.5400. Load Mbps    D  r  o  p  p   i  n  g   P  r  o   b .   (   %   )   No MLBPRB utilisation-based MLBLoad-based MLB Figure 3 Dropping probability as a function of offered load. 11.522.533.5400.511.522.533.544.55Offered Load Mbps    H  a  n   d  o  v  e  r   f  a   i   l  u  r  e  p  r  o   b .   (   %   )   No MLBPRB utilisation-based MLBLoad-based MLB Figure 4 Handover failure probability as a function of offered load.
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!